Level set tracking using shape prior and global-local color model

This paper presents a new level set tracking method for the tracking of non-rigid objects in videos. The main idea is to introduce a new probabilistic model to classify the background and object more robustly, which is then encoded as a new term for the energy function for solving the level set problem. Also, we propose to use bounding box of the contour as a shape prior, which simplifies the contour tracking process. Another contribution is to design temporal smoothness term. Experimental results shows that our method performs well on several challenging sequences in comparison with conventional algorithms.